Microsoft's 37-page "humanist AI code of conduct" puts people over AI

Microsoft's code arrives after a summer of unsettling AI-agent incidents, most notably the OpenAI / Hugging Face episode in which a swarm of agents attacked targets and hacked into the "grader" evaluating their performance, as well as a "wiki incident" where out-of-control agents hijacked a German wiki site. Those events pushed researchers to call for slower AI model progress, with Anthropic CEO Dario Amodei urging a coordinated slowdown over the weekend. Microsoft is positioning its 37-page "humanist AI code of conduct" as a direct answer to that anxiety, with AI CEO Mustafa Suleyman calling Anthropic's speculation "really, really dangerous" during a June Decoder episode.

The code also draws lines around model transparency: Microsoft commits that its models won't communicate beyond simple human understanding and should fail a task rather than violate its rules. That follows researcher concerns this month that OpenAI's GPT-6 Astra model reveals less of its reasoning than other AI models. Microsoft CEO Satya Nadella and OpenAI CEO Sam Altman have both endorsed human control and pacing, though Altman said slowing is not the same as stopping development.

The test is likely to be whether Microsoft's commitments hold as it works to compete with Google, Anthropic and OpenAI. Microsoft isn't one of the top AI providers yet, and Suleyman has said the goal is to become one of the top four labs in the world. Whether the code constrains its models in practice, or simply differentiates them, will depend on how the company evaluates real-world performance and emotional dependence in its partners and products.